Systems Requirement: 2021-2022 Competition

Task Overview

Task 1

  • Provided images and video of the building and the area within 50m of the building. Intelligence about delivery of a device including Device size, weight and shape. Device will be rigid, no larger than 20cm on any one axis.

  • Device may be close as 1m to a vertical structure.

  • Must detect device, pick it up and transport to secure drop off location.

  • Marked on: Ability to detect device, delivery of the device, time

Task 2

  • Provided a QR code. Contains

    • Questions

    • Date, Time

    • Device and Sensor ID

    • Longtitude and Lattitude

  • Fly to intrusion location

  • Detect and track intruder over path, camera must be fixed on intruders and viewable at GCS. Determine key questions to answer the questions given. Provide a copy of the video to judges.

  • Create a trace on the provided map showing the path of the intruders.

  • Marked on: Correctly interpreting intrusion message, automatic launch and flight to location, detection and identification of intruder, answers to questions, map trace

Requirements Analysis: Clarify which requirements we want to pursue based on how we could pursue it.

Requirement: How it contributes to marking

High-level implementation: Details

High-level Implementation: Modules used

Likelihood of success / Finalization

Next steps to confirmation

Requirement: How it contributes to marking

High-level implementation: Details

High-level Implementation: Modules used

Likelihood of success / Finalization

Next steps to confirmation

Device detection (Task 1): Enhances our ability to detect device.

Fairly difficult, and maybe we should deprioritize it.

Point contribution: indirect

Image segmentation:

  1. Perform image segmentation by edge detection to create bounding boxes.

  2. Filter out bounding boxes that fit size of target.

  3. Map targets.

  4. Send frames with bounding boxes to ground alongside votes.

TargetAcqusition

TargetMapping

Decklinksrc

GroundSend

unsure

research only

low priority

Is there a contrast between the device and the ground?
Is there anything else on the ground?
What does the building look like? Is it actually a wall? How tall?

Device pickup (Task 1): Autonomous pickup

Mechnical and Firmware might need our help

  1. Drive the UAV towards it autonomously by comparing the world position of target and world position of drone.

  2. Pickup when the UAV is in the air.

  3. grab the device tightly

TBD

unsure

research only

low priority

Talk to Mech and FW and see how to implement this step

 

Device delivery (Task 1):

Enhances our ability to drop off the device precisely

Point contribution: Direct

  1. Drive to the drop off location (coordinates?)

  2. enable camera and align the UAV against the barrel (87cm high and 58 cm in diameter)

  3. lower the UAV and drop off gently (device is explosive)

Can reuse device pickup

TBD

unsure

research only

low priority

Ask if coordinates of the drop off location is provided

QR Scan (Task 2): Allows for autonomous takeoff and message reading

Point contribution: direct

  1. Scan QR code using existing implementation.

  2. Send message to ground station for human read.

  3. Extract longtitude and lattitude

  4. Send to zeropilot

QRScanner

FltConnSend

GroundSend

Likely

finalized

high priority

Ask FW about auto-navigation using lat/long
Ask about QR code position

Detect and highlight intruder (Task 2): Marginally improves detection of intruder

Point contribution: indirect

  1. Train YOLOv5 model on person from above.

    1. Note, this requires us to be able to change how TargetAcquisition is implemented on the fly (i.e different TargetAcquisition modules).

      1. Resolution: Keep it at two classes and switch in program.

    2. During future comps, just train on multiple classes for multi object detection.

    3. Make sure training images are of people from above

  2. Perform object detection on frames to generate bounding boxes.

  3. Send frame with bounding box to ground.

TargetAcquistion

DecklinkSrc

GroundSend

 

Likely

finalized

med priority

 

Provide control input to autonomously track intruder (Task 2): Improves tracking of intruder.

Alternate idea: Move camera gimbal to keep person in centre.

Point contribution: indirect

  1. Use detection of intruder to find bounding box centoid

  2. Convert bounding box centoid to world coordinate

  3. Send world coordinate to zeropilot, which can compare target coordinate and drone position to perform control inputs.

TargetAcquisition

DecklinkSrc

Geolocation

FltConnSend

possible

unfinalized

med priority

Coordination with FW for support on ZP side. (This could be a problem and is quite difficult, limiting factor)
Confirm what the intruders behaviour will look like (run/walk/wave arms etc.)
Confirm with Mech/FW/EE that we are definitely going to support hover.
Obstacles on flight area?

Visual trace of intruders (Task 2): Improves path tracking of intruder.

Seperate since we’re just sending it to the ground.

Point contribution: indirect

  1. Use world coordinates of target from intruder tracking.

  2. Store world coordinates throughout tracking.

  3. Plot coordinates on map as path.

  4. Send down to ground station.

TargetAcquisition

DecklinkSrc

Geolocation

GroundSend

possible

unfinalized

med priority

 

Autonomously answering questions

Detect? ML??

 

Unlikely

unfinalized

low priority

Ask what kind of questions? (Clothes etc.)

Likelihood of Success Options:

  • UNSURE : Research required to determine if possible and difficulty

  • Unlikely : Implementation that has never been done or scarcely documented, inside or outside WARG.

  • possible : Implementation has been done before with some documentation, or well documented but technically difficult.

  • Likely : Implementation has been done before, partially or completely, with extensive documentation and/or code, or some documentation but technically simple.

Point Contribution:

  • indirect : This feature is not a part of the marking rubric directly, but it enhances our team ability and contributes points to the ingenuity of our solution, or improves our ability but is not strictly necessary to achieve a certain rubric.

  • direct : This feature is a direct part of the marking rubric.

Technical Notes

  • Qt has a map plugin and video/media player plugin so we could hypothetically add these.

  • Qt doesn’t seem to have any performance issues.

  • For multiple TX, just have them all transmitting on the same channel.